lostanlen / ismir2016

Deep Convolutional Networks on the Pitch Spiral for Musical Instrument Recognition
42 stars 11 forks source link

Environment setup #2

Open mc0514 opened 8 years ago

mc0514 commented 8 years ago

Hi, could you show me how to run your code?

lostanlen commented 8 years ago

I hope this helps.

Kaustubhk12 commented 7 years ago

Hi, I tried running your code but I am unable to install medleydb.sql package required by singlelabel.py, it has been deprecated.

Kaustubhk12 commented 7 years ago

Any suggestions as to how to run your code inspite of the fact that medleydb.sql is not available/ deprecated and your code uses medleydb.sql?

scriptator commented 7 years ago
In [1]: import DeepInstruments
Using TensorFlow backend.
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-1-fda0621afab2> in <module>()
----> 1 import DeepInstruments

/Users/johannesvass/ownCloud/Studium/2017S_Bachelorarbeit/ismir2016/DeepInstruments/__init__.py in <module>()
     51 import DeepInstruments.audio
     52 import DeepInstruments.descriptors
---> 53 import DeepInstruments.learning
     54 import DeepInstruments.singlelabel
     55 import DeepInstruments.symbolic

/Users/johannesvass/ownCloud/Studium/2017S_Bachelorarbeit/ismir2016/DeepInstruments/learning.py in <module>()
----> 1 from keras.models import Graph
      2 from keras.layers.advanced_activations import LeakyReLU
      3 from keras.layers.core import Dense, Dropout, Flatten
      4 from keras.layers.convolutional import Convolution2D, MaxPooling2D
      5

ImportError: cannot import name 'Graph'

Also, keras.models.Graph seems to be deprecated, see https://github.com/fchollet/keras/issues/4661

lostanlen commented 6 years ago

In the matter of one year, this code has gone completely stale. This is because Keras 2 has removed the Graph object (now uses a "functional API") and also because MedleyDB has removed its SQL backend. What is your use case? If what you want is to re-use the spiral convolutional network for your own dataset/task, I suggest you re-implement it from scratch according to the code which is in learning.py. I can help you with putting it up to date for Keras v2. The rest of the package is mostly data wrangling related to MedleyDB.